Description of problem:

old way with yum-utils: repoquery --requires --resolve <some_pkg>
new way with dnf-plugins-core: dnf repoquery --requires --resolve <some_pkg>

Version-Release number of selected component (if applicable):

fedora 22

% rpm -qa | egrep 'yum-utils|dnf-plugins-core'
dnf-plugins-core-0.1.15-1.fc22.noarch
yum-utils-1.1.31-508.fc22.noarch
python-dnf-plugins-core-0.1.15-1.fc22.noarch


How reproducible:

always

Steps to Reproduce:

Compare wall time to run yum-utils' repoquery vs. dnf-plugins-core:

example:

sh -c 'time repoquery --requires --resolve gscan2pdf'
  .
  .
real    0m4.316s
user    0m4.064s
sys     0m0.249s

sh -c 'time dnf repoquery --requires --resolve gscan2pdf'
  .
  .
real    4m7.166s
user    4m3.422s
sys     0m3.731s

On both cases, the metadata cache was up to date before running the repoquery.

Without --resolve, the times are comparable (< 5 s).

Is this a difference due to fundamentally different issues between yum & dnf?  The magnitude of speed difference seems to be indicate than just 'python adds a little overhead'.

Hints to where to start profiling are appreciated - I currently have near zero internals knowledge for yum/dnf, but I'm willing to dig a little.